Cross-Category Item Associations Using Machine Learning

1. A system comprising: one or more processors; and memory storing computer-executable instructions that, when executed, cause the one or more processors to perform acts comprising: creating training data by: analyzing training image data and training text data associated with each of a plurality of training items; assigning an item category for each of the plurality of training items; identifying, for each of the plurality of training items, from the training image data and the training text data, attributes of each of the plurality of training items, wherein the attributes are associated with a visual appearance of each of the plurality of training items; identifying parameter types for each of the item categories, wherein the item categories comprise one or more of color and pattern; and categorizing each of the attributes for each of the plurality of training items into at least one of the parameter types; creating an association model using a machine learning engine by: analyzing the training image data, the training text data, the item category, the attributes, and the parameter types to create correlations between the training image data, the training text data, the item category, the attributes, and the parameter types; and creating the association model that uses the attributes of the parameter types to calculate a first association score based on the correlations, the association model including weights for individual parameter types that apportion an influence of the parameter types in calculation of the association score; and deploying the association model by: receiving first image data that depicts a first item in a first item category and first text data associated with the first item; analyzing the first image data and the first text data to identify, from the first image data and the first text data, first attributes of the first item, wherein the first attributes are associated with a first visual appearance of the first item; categorizing the first attributes into first parameter types associated with the first item category, wherein the first parameter types comprise one or more of color and pattern; searching, within a second item category that is different than the first item category and based on the first attributes or the first parameter types, for a second item having a second visual appearance that is similar to the first visual appearance of the first item; identifying the second item from a plurality of second items that are different than the first item and that are within the second item category, wherein identifying the second item comprises: calculating, using the association model, a second association score associated with second attributes of second parameter types associated with each of the plurality of second items to determine similarities between the first attributes associated with the first item and the second attributes associated with each of the plurality of second items; and identifying the second item having the second visual appearance similar to the first visual appearance of the first item from the plurality of second items for which the second association score is at least one of above a threshold association score or among highest ranked association scores.

2. The system as recited in claim 1 , wherein the acts further comprise updating the weights based at least in part on repeated calculations of the first association score to improve an accuracy of the association model.

3. The system as recited in claim 1 , wherein the acts further comprise providing a user interface configured to receive user designations that modify the weights for at least one of the parameter types.

4. The system as recited in claim 1 , wherein the first item category comprises at least one of first clothing or first shoes for a first person having a first size, and the second item category comprises at least one of second clothing or second shoes for a second person having a second size different than the first size, and wherein the first item is provided by a first provider and the second item is provided by a second provider different than the first provider.

5. The system as recited in claim 1 , wherein the first item category is women's clothing and the second item category is girls' clothing, and wherein the parameter types further comprise at least one of clothing type, clothing style, clothing shape, neckline type, sleeve type, or hemline type.

6. A method comprising: analyzing first image data depicting a first item in a first item category and first text data associated with the first item; identifying, from the first image data and the first text data, one or more first attributes of the first item, wherein the one or more first attributes are associated with a first visual appearance of the first item, and wherein identifying the one or more first attributes comprises extracting at least one of color data or pattern data from at least one of the first image data or the first text data; categorizing the one or more first attributes into one or more first parameter types associated with the first item category, wherein the one or more first parameter types comprises at least one of color or pattern; searching, within a second item category that is different than the first item category and based on the one or more first attributes or the one or more first parameter types, for a second item having a second visual appearance that is similar to the first visual appearance of the first item; and identifying the second item from a plurality of second items that are different than the first item and that are within the second item category, wherein identifying the second item comprises: calculating, using an association model, an association score associated with one or more second attributes of one or more second parameter types associated with a subset of the plurality of second items to determine one or more similarities between the one or more first attributes associated with the first item and the one or more second attributes associated with the subset of the plurality of second items; and identifying the second item having the second visual appearance similar to the first visual appearance of the first item from the plurality of second items for which the association score is at least one of above a threshold association score or among highest ranked association scores.

7. The method as recited in claim 6 , further comprising assigning weights to a second subset of the one or more second parameter types, each of the weights apportioning an influence of the second subset of the one or more second parameter types on the association score.

8. The method as recited in claim 7 , wherein assigning the weights to the second subset of the one or more second parameter types comprises providing a user interface configured to receive user designations for one or more weights for at least one of the one or more second parameter types.

9. The method as recited in claim 6 , further comprising calculating a confidence score associated with each of the plurality of second items identified as having at least one of a second association score that is above the threshold association score or that is among the highest ranked association scores, the confidence score being associated with a probability that the association score is accurate.

10. The method as recited in claim 6 , wherein the first item category is first clothing for a first person provided by a first provider, and the second item category is second clothing for a second person provided by a second provider.

11. The method as recited in claim 6 , wherein the first item category is women's clothing and the second item category is girls' clothing, and wherein the parameter types comprise at least one of clothing type, clothing style, clothing shape, neckline type, sleeve type, or hemline type.

12. The method as recited in claim 6 , wherein categorizing the one or more first attributes into the one or more first parameter types comprises extracting image features from the first image data and text features from the first text data, and associating a second subset of the image features and the text features with a parameter type.

13. The method as recited in claim 6 , wherein calculating the association score associated with the one or more similarities comprises calculating a parameter association score for each parameter type, and combining the parameter association score for each parameter type to obtain the association score.

14. The method as recited in claim 6 , further comprising training the machine learning engine, wherein the training comprises: analyzing, using the machine learning engine, training image data and training text data associated with a second subset of a plurality of training items; assigning an item category for the second subset of the plurality of training items; identifying, for the second subset of the training items, from the training image data and the training text data, attributes of the second subset of the plurality of training items; identifying parameter types for a third subset of the item categories; and categorizing each of the attributes for the second subset of the plurality of training items into at least one of the parameter types.

15. The method as recited in claim 14 , further comprising creating training data providing correlations between the training image data, the training text data, the item category, the attributes, and the parameter types.

16. The method as recited in claim 6 , further comprising creating the association model, based at least in part on training data, for determining the one or more similarities between the one or more first attributes of the first item and the one or more second attributes associated with a second subset of the plurality of second items in the second item category.

17. The method as recited in claim 6 , further comprising outputting a user interface to receive at least a first user selection of the first item and a second user selection of a second category associated with the plurality of second items.

18. A system comprising: one or more processors; and memory storing computer-executable instructions that, when executed, cause the one or more processors to perform acts comprising: identifying, from first image data and first text data, one or more first attributes of a first item in a first item category, wherein the one or more first attributes are associated with a first visual appearance of the first item, and wherein identifying the one or more first attributes comprises extracting at least one of color data or pattern data from at least one of the first image data or the first text data; categorizing the one or more first attributes into one or more first parameter types associated with a first item category, wherein the one or more first parameter types comprises at least one of color or pattern; searching, within a second item category that is different than the first item category and based on the one or more first attributes or the one or more first parameter types, for a second item having a second visual appearance that is similar to the first visual appearance of the first item; and identifying, using a machine learning model, the second item from a plurality of second items that are different than the first item and that are within the second item category, wherein identifying the second item comprises: calculating, using a machine learning model, a similarity score associated with one or more second attributes of one or more second parameter types associated with a subset of the plurality of second items to determine one or more similarities between the one or more first attributes associated with the first item and the one or more second attributes associated with the subset of the plurality of second items; and identifying the second item having a second visual appearance similar to the first visual appearance of the first item for which the similarity score is at least one of above a threshold similarity score or among highest ranked similarity scores.

19. The system as recited in claim 18 , wherein the parameter types comprise at least one of clothing type similarity, clothing style similarity, clothing shape similarity, neckline type similarity, sleeve type similarity, or hemline type similarity.

20. The system as recited in claim 18 , wherein calculating the similarity score comprises calculating a parameter type score for each parameter type, and combining the parameter type score for each parameter type to generate the similarity score.